includeInteraction: Function to create user-specified interactions for a Siena...
In RSiena: Siena - Simulation Investigation for Empirical Network Analysis

includeInteraction

R Documentation

Function to create user-specified interactions for a Siena model.

Description

This function allows the user to include or exclude an interaction effect in a Siena effects object.

Usage

includeInteraction(myeff, ..., include = TRUE, name = myeff$name[1],
    type = "eval", interaction1 = rep("", 3), interaction2 = rep("", 3),
    fix=FALSE, test=FALSE, random=FALSE,
    initialValue=0,
    character = FALSE, verbose = TRUE)

Arguments

`myeff`	a Siena effects object as created by `getEffects`
`...`	2 or 3 short names to identify the effects which should be interacted.
`include`	Boolean. default TRUE, but can be switched to FALSE to turn off an interaction.
`name`	Name of dependent variable (network or behavior) for which interactions are being defined. Defaults to the first in the effects object.
`type`	Type of effects to be interacted.
`interaction1`	Vector of Siena objects where needed to completely identify the effect e.g. covariate name or behavior variable name. Trailing blanks may be omitted.
`interaction2`	Vector of siena objects where needed to completely identify the effect e.g. covariate name or behavior variable name. Trailing blanks may be omitted.
`fix`	Boolean. Are the effects to be fixed at the value stored in `myeff$initialValue` or not.
`test`	Boolean. Are the effects to be tested or not (requires `fix`).
`random`	For specifying that the interaction effect will vary randomly; not relevant for `RSiena` at this moment. Boolean required. Default FALSE.
`initialValue`	Initial value for estimation. Default 0.
`character`	Boolean: are the effect names character strings or not.
`verbose`	Boolean: should the print of altered effects be produced.

Details

The details provided should uniquely identify up to three effects. If so, an interaction effect will be created and included or not in the model.
Whether interactions between two or three given effects can be created depends on their interactionType (which can be, for dependent network variables, empty, ego, or dyadic; and for dependent behavioral variables, empty or OK). Consult the section on Interaction Effects in the manual for this. The interactionType is shown in the list of effects obtained from the function effectsDocumentation.
The short names must not be set between quotes, unless you use character=TRUE.
From the point of view of model building it is usually advisable, when including an interaction effect in a model, also to include the corresponding main effects. This is however not enforced by includeInteraction().

As from version 1.3.24, effects object have a "version" attribute. Effects objects including interaction effects are not necessarily compatible between versions of RSiena. Therefore it is recommended to create such effects objects again when changing to a new version of RSiena. If an effects object including any interaction effects is used from an old version of RSiena, this will lead to a warning when running siena07.

An interaction effect does not have its own internal effect parameter. The internal effect parameters of the interacting main effects are used, whether or not these are included in the model. This implies that if an interaction effect is included but not the corresponding main effects, or not all of them, then nevertheless the internal effect parameters as specified in the effects object are used for the interaction. These can be set using function setEffect with the desired value of parameter and (in this case) include=FALSE.
The values of the internal effect parameters can be checked for a sienaFit object ans produced by siena07 by looking at ans$effects, which is the requested effects object to which the main effects of the user-defined interactions were added, if they were not yet included.

Interaction effects are constructed from effects with shortName unspInt (for networks) and behUnspInt (for behavior) by specifying their elements effect1 and effect2, and possibly effect3. The shortName is not altered by this function.

The number of possible user-specified interaction effects is limited by the parameters nintn (for dependent network variables) and behNintn (for dependent behavior variables) in the call of getEffects, which determine the numbers of effects with shortNames unspInt and behUnspInt.

The input names interaction1 and interaction2 do not themselves refer to created interactions, but to dependence of the base effects on other variables in the data set. They are used to completely identify the effects.

Further information about Siena effects objects is given in the help page for getEffects.

A list of all effects in a given effects object (e.g., myeff), including their names of dependent variables, effect names, short names, and values of interaction1 and interaction2 (if any), is obtained by executing effectsDocumentation(myeff).

Value

An updated version of the input effects object; if include, containing the interaction effect between "effect1" and "effect2" and possibly "effect3"; if not, without this interaction effect. The shortName of the interaction effect is "unspInt" for network effects and "behUnspInt" for behavior effects.
If verbose=TRUE, details of the fields altered will be printed.

Author(s)

Ruth Ripley, Tom Snijders

References

See https://www.stats.ox.ac.uk/~snijders/siena/

Examples

mynet <- sienaDependent(array(c(s501, s502, s503), dim=c(50, 50, 3)))
alc <- varCovar(s50a)
mydata <- sienaDataCreate(mynet, alc)
myeff <- getEffects(mydata)
myeff <- includeEffects(myeff, transTrip)
myeff <- includeInteraction(myeff, recip, inPop)
myeff <- includeEffects(myeff, egoX, altX, simX, interaction1="alc")
myeff <- includeInteraction(myeff, recip, simX, interaction1=c("", "alc"))
myeff

RSiena documentation built on June 22, 2024, 11:05 a.m.